Circadian rhythms and mood: Opportunities for multi-level analyses in genomics and neuroscience

Functional plasticity: A good clock is reliable and adjustable

Circadian regulation is a quintessential example of functional plasticity, as it demonstrates the balance between consistency and flexibility. The baseline rhythms, particularly in the SCN, are robust to genetic and environmental perturbations 6, 7. However, these rhythms can adapt to major changes in the operating environment, including seasonal changes of day length, and, in the case of modern humans and experimental animals, sudden changes in daily light schedule 5, 8. Evolution has enabled the circadian system to respond to gradual seasonal changes. However, rapid phase resetting as demanded by air travel or shift work is an evolutionarily recent, novel stressor; and the mammalian circadian system is capable of responding to such acute demands. Even so, while most people can make the transition to a new schedule, frequent shifts can lead to long-term health consequences, such as increased risk of cardiovascular and immune dysfunction.

Network properties: No clock works alone in the circadian system

In the SCN, rhythms are maintained as both single-cell and ensemble properties. Each SCN neuron generates intrinsic rhythmicity through interlocking feedback loops involving a set of “core clock genes” and their protein products. Cell-autonomous rhythms are synchronized at the level of cell populations. The SCN exerts top-down control over other brain regions and peripheral oscillators while receiving feedback from them. Thus, the circadian control system is a hierarchical network, involving “elementary clocks” in individual cells, “ensemble clocks” in cell populations, coupling between central and peripheral oscillators, and interactions with external input 9 (Fig. 1). As an analogy, different oscillators behave like sections in an orchestra – within each section every member (the cell) is a self-sufficient musician, and every musician plays in ensemble by listening to the others and following a rhythmic input (the daily cycle of environmental light). Each section has its distinct role, but harmonizes with the other sections via a precise phase relationship. There exists a standard temporal protocol, but it is possible to vary it or to reset the phase relationship. Healthy operation of the system relies on both central control and local autonomy, and the right level of plasticity. Compared to isolated cells, ensemble rhythms of the network are more resistant to genetic or environmental perturbations 6, 7, 10, serving as a reminder that circuit-level properties are central to any effort to understand brain function.

Neural dynamics: The circadian system provides an ideal model to study temporal organization

In the SCN, adaptation to seasonal changes of day length or temperature is achieved not by changed oscillating properties of individual neurons but by the altered phase relationship among them 11, 12. Circadian regulation thus provides an ideal model to study the role of temporal organization. The master-subordinate relationship between the SCN and other oscillators is known; so is the core set of 10–20 genes and how they interact with each other to maintain cell-autonomous rhythms. The next several hundred genes in specific tissues have begun to be recognized in genomic studies. There is a wealth of existing data on the wiring diagram and connection parameters for both intra- and intercellular interactions 13. The dynamic behavior of circadian oscillators has been extensively modeled (e.g. 14), allowing quantitative understanding of coupling strength, phase distribution, and their consequences 15. In the example of recovery from jet lag, one can ask some modestly ambitious questions as a starting point for understanding mechanisms of resynchronization and stability: Since the dorsal and ventral regions of the SCN show different response speeds to light schedule changes 16, how do the two circuits regain phase cohesion? How does the tissue ensemble transition from one stable state to another? Can the mechanisms of phase resetting teach us something about neural regulation in other systems?

Relevance to other systems: Everything happens in its own time, for a reason, and together

Adaptation in other mental functions, including mood, is also best understood in terms of network properties, including intrinsic dynamics, responses to stress, and alternative (meta)stable states. This essay is partly inspired by the premise that the manner in which the brain manages the synchrony and asynchrony of neuronal populations is crucial to its normal operation. Circadian rhythm regulation is not the only example where temporal patterns in neural circuits – rather than constitutional features such as static levels of neurotransmitters – encode the function of the system and mark the divide between health and disease 17, 18. In animal models, short-term dynamic modulation of synaptic plasticity is correlated with behavior in a working memory task 19. Some cardinal symptoms of depression and schizophrenia are linked to aberrant timing in information processing 20. These results strongly suggest that inter-circuit coupling and timing could be a universal mechanism for network-level function, and that impaired temporal organization could manifest as behavioral deficits 21. Further, since maladaptation may arise from an inability to regain proper synchrony, this network-level phenotype can be a target for therapy; for example, clinical data show that social rhythm stabilization can improve recovery from bipolar depression 22, 23.

Circadian rhythm and mood regulation share a similar temporal scale. The adjustment to jet lag takes many days to harmonize across oscillators, despite the fact that ventral SCN itself is entrained quickly 16. The characteristic tempo of this form of inter-circuit plasticity, spanning days to weeks, is reminiscent of the mood cycling time in bipolar disorders (BD) and the therapeutic delay observed for classic antidepressants, which take 2–4 weeks to work fully (in the case that they are effective for that individual). Could circadian phase flexibility share a common mechanism with mental resilience 24? Do patients with depression exhibit different patterns of synchronicity between oscillators compared to healthy subjects, reflecting too much or too little plasticity? Can we discover causal network features for a brain disease as opposed to causal molecular or cellular features?

Mendelian and common forms of sleep disorders cover a spectrum of genetic effects

Several Mendelian forms of sleep disorders have been described 45. A pedigree of familial advanced sleep phase disorder (FASPD) carries a S662G variant in hPer2, a human homolog of the period gene in Drosophila, a core clock gene 46. When tested in transgenic mice, the S662G change recapitulated the human phenotype 47. Similar discoveries were subsequently made for other clock genes in additional pedigrees of sleep disorders 48, 49. Meanwhile, mouse models of Clock, Npas2, Bmal1, Cry1, and Cry2 show altered sleeping traits (reviewed in 45). Collectively, these studies provided convincing support for the importance of core clock genes in regulating sleep behavior. Notably, the loss of a single clock gene in animal models is often insufficient to disrupt normal rhythms unless environmental time cues are also disrupted. In the normal environment, multiple genetic defects are often required to elicit altered behavior 10, demonstrating robustness against genetic perturbations.

In humans, the prevalence of Mendelian sleep disorders is not accurately known, but they are likely rare. The population burden of common sleep disorders is yet to be defined. Twin studies have shown that sleeping traits have a moderate genetic component in the general population, with narrow-sense heritability in the 30–40% range 50-53. Candidate gene-based association studies of common circadian rhythm disorders have focused primarily on known clock genes and have been reviewed previously 45.

Unlike candidate gene studies, genome-wide association studies (GWAS) enable an unbiased search beyond known candidates. To date, three GWAS have been conducted for sleeping traits, and have identified association signals near the NPSR1 and PDE4D genes 54, in CACNA1C, encoding an L-type calcium channel 55, and ABCC9, encoding an ATP-sensitive potassium channel 56. Attempts to replicate the CACNA1C signal have produced mixed results 55, 57. Notably, none of the core clock genes reached genome-wide significance in these GWAS, suggesting that, with the sample size currently available, the association between common variants in known clock genes and common circadian rhythm traits have not yielded replicable results.

GWAS for psychiatric disorders: Large community efforts, variable yields

An extensive literature exists regarding linkage analyses and candidate gene association studies of psychiatric traits. Taken together, candidate gene-based association for complex traits has a lackluster history, marked by poor reproducibility 58, 59. Here, I will focus on GWAS, in which the statistical threshold is higher than for candidate genes, but the signals that did achieve genome-wide significance have proven to be more reproducible. Dozens of GWAS have been conducted for five mental disorders: MDD, BD, schizophrenia, autism, and attention deficit hyperactivity disorder; and many more GWAS are added every year. A meta-analysis of MDD (11,215 patients and 9,761 controls) did not identify risk loci with genome-wide significance 60. A similar analysis of BD (11,974 patients and 51,792 controls) identified CACNA1C, a calcium channel, and ODZ4, encoding a cell surface protein 61. A meta-analysis of schizophrenia (n > 50,000 patients and controls) implicated 22 loci 62, 63. Pathway analysis showed that calcium channels are enriched in BD association intervals. Of note, the CACNA1 gene was among the top findings in the GWAS for sleeping traits described above 55, suggesting a potential common genetic link between sleep and mood disorders.

The polygenic nature of psychiatric disorders brings new challenges

A meta-analysis of GWAS for schizophrenia suggested a polygenic model, involving common alleles across thousands of loci, each contributing very small effects 64. Since then, a polygenic genetic architecture has been similarly reported for other complex traits such as multiple sclerosis and body mass index 65, 66. Combined analyses of the five psychiatric diseases mentioned above, including 33,332 cases and 27,888 controls, demonstrated strong cross-disorder genetic overlap 67, 68. Four regions, including CACNA1C, were associated with all five disorders. Among MDD, BD, and schizophrenia, risk scores calculated from polygenic signals in one disorder can explain a statistically significant (albeit small) portion of the variance of a second disorder, particularly between BD and schizophrenia, suggesting shared genetic etiology spreading numerous loci. The latest analysis of schizophrenia implicated 6,300–13,200 independent SNPs 63.

The findings described above have provided deeper insight into the genetic basis of psychiatric disorders, but they have also brought us head-on into the next puzzle: How to make use of these findings? The polygenic scores, while compelling in a statistical sense, are remarkably diffuse in a biological sense. The sheer number of risk factors has no parallel in the history of epidemiology. For a study cohort, the inherited risks are apparently spread over thousands of loci and subtly implicate numerous genes and pathways. However, for an individual patient, does he/she come to develop the disease by equally dispersed biological routes? Or, could the polygenic signal be merely illusory, formed by pooling large numbers of patients with heterogeneous causes, each of whom develop a disease due to far fewer variants, but of higher phenotypic impact 69? If the latter is true, what is the best way to identify the heterogeneous, patient-specific genetic factors? Questions such as these are the subject of ongoing debate and are discussed further in Box 1.

Rare, high-impact variants take center stage

To date, genetic studies have identified ABCC9 and possibly CACNA1C for common sleeping traits, CACNA1C and ODZ4 for BD, no significant signals for MDD, 22 associated regions for schizophrenia, and broad, polygenic signals shared across multiple disorders. These GWAS, however, have not identified any core clock genes with genome-wide significance. Taken together, we currently have very few specific DNA variants with large effects to motivate experimental follow-up, especially at the interface of circadian rhythm and mood regulation. Polygenic signals and most pathway findings carry limited specificity. For example, when a psychiatric disease is said to involve hundreds of genes that, as a whole, are enriched for neurodevelopment or synaptic transmission, the notion lies closer to the question than to the answer, as it is hardly a novel insight that brain disorders will likely involve neurodevelopment and neural transmission. In the absence of a credible molecular target, these results also do not lead to clearly falsifiable predictions.

In the near future, given the paucity of unequivocal leads, it is profitable to focus on de novo, Mendelian or nearly Mendelian cases, using rare DNA variants of large impact as the source of informative genetic perturbation. The increased availability of DNA variation data in the general population, produced by large-scale DNA sequencing in studies of other human diseases, opens up opportunities of finding genetic signals of testable consequences. For example, it would be quite interesting to ask if carriers of loss-of-function mutations in core clock genes present altered sleep habits or some forms of previously unrecognized sleep disorders. If the answer is no, the next interesting question is how molecular defects in key clock genes could be compensated at the physiological level. If the answer is yes, the identified individuals, with reverse-ascertained circadian abnormalities, would provide exciting new models to probe the molecular interactions underlying circadian regulation. Logistically, this genotype-looking-for-phenotype approach will be greatly aided by a national or international registry of genetic variants of unknown functional significance. When call-back phenotyping becomes feasible, it could extend from sleep behavior to mental function parameters, providing the much needed epidemiological data to address the chicken-versus-egg question.

How might this integration take place?

To illustrate how all the tools and data can be marshaled under an integrative “framework”, I will use a hypothetical scenario. Let us suppose that, first, genetic or gene expression data have implicated a calcium channel gene in mood disorders. And second, screening of Mendelian families or de novo cases has identified rare mutations with a strong predicted impact on channel function. Using this genetic lead, we can screen the general population to identify additional subjects carrying high-impact mutations in this gene, and in the meantime, develop targeted animal models such as gene-disrupting transgenics or optogenetic models. Using these human subjects or animal models, we can pursue the pathophysiological relevance of the mutations at multiple levels simultaneously (see Box 2). While none of the experiments in Box 2 addresses the complete mechanism, the multi-level data from the same sample series have the best chance to “fill in” the succession of gaps from genes to behavior. And the experiments, when driven by concrete genetic hypotheses, have the best chance to unravel the causal relationships in the system.

Because circadian regulation represents a uniquely workable system, it is suitable as the testing ground of this integrative research program. For example, coupling between the SCN and other oscillators should be contrasted between the mutation carriers and normal controls, especially in the context of dynamic network properties. Specifically, since the master oscillator resides in a well-defined anatomical unit, the SCN, a valuable focal point of research is to interrogate the strength and phase of functional coupling between the SCN and other brain oscillators 83, 84, such as the locus coeruleus, which regulates the transition from attention to behavioral flexibility 85, or amygdala, which regulates fear, anxiety, and depression symptoms. These experiments will directly address if, and how, a single genetic mutation can affect small-circuit plasticity, temporal coordination among networks, and behavioral outcome 16.

In practice, the best efforts of cross-level integration may break down. For example, despite clear evidence for the involvement of the monoamine neurotransmitters at the cellular and pharmacological level, PET studies could not correlate depression severity to the distribution and occupancy of monoamine receptors/transporters 86. In animal models, an engineered genetic defect often fails to impinge on cellular or circuit phenotypes because of (presumably) compensatory effects of other genes. Sometimes, it takes two genetic defects to elicit a disease-like condition, but only in a certain genetic background or under certain environmental stress. In this regard, circadian regulation and its role in mood disorders present the opportunity to clarify the power and limitations of multi-scale integration.

Three hurdles: Developmental timing, nonlinear system, and ontological mismatch

The difficulty of translating genetic findings to physiological explanations is rooted in the unexplained coupling between different levels of biological organization, and the critical dimension of time, involving both developmental timing and network dynamics (Fig. 4). Neural outputs such as emotion, cognition, and behavior are not directly coded, or understood, by properties of single cells or bulk tissue averages. Rather, the brain works as spatially distributed hierarchical neural networks, molded by an individual's genetics, early development, and life experience. If much of the initial impairment takes place during a critical period of brain development, as is plausible for BD and schizophrenia, adult-onset symptoms may merely occur as aftershocks of an unobserved earlier distress. The ideal strategy for probing the temporal sequence of brain disorders is to collect the longitudinal series that covers the relevant episodes, but this is always a daunting task in human studies. In short, we still lack suitable data to leap this hurdle.

An even greater challenge lies in understanding the iterative, nonlinear interactions between genes and environment. A long-standing riddle in psychiatric illnesses is the contrast between their diverse plausible causes and the accelerated appearance of distinctive symptoms. This contrast echoes the deeper contradiction that knowledge in epidemiology is statistical, yet the practice of medicine is personal. From the statistical point of view, genetic predisposition in most cases appears to be thinly distributed over thousands of loci 63. Similarly, neural deficiencies in any population of patients implicate a wide mix of neurotransmitter systems and brain regions. However, at the individual level, the clinical presentations are clustered, distinctly recognizable, and often strongly familial, suggesting a restricted number of archetypal disturbances, possibly involving a small number of genetic origins, neurotransmitters, and brain circuits. This implies that in the natural history of a given patient, the interactions between his/her unique genes and unique environment must be dynamic, convergent, self-organizing, and self-reinforcing, bearing resemblance to the classic idea of canalization 87, 88. New analytical algorithms are needed to capture such a clearly nonlinear system, as the current case-control studies are primarily designed to screen for simple, stable effects shared across a large cohort.

Lastly, we face an ontological mismatch between the working concepts of different disciplines. As an analogy, something as simple as water has many levels of understanding: The structure of H₂O does not automatically explain its taste, wetness, or how it makes the floor slippery. Likewise, the fundamental units of genetic analysis – genes, their expression patterns in time and space, and pathways – remain to be aligned with the fundamental units of neuroscience: cells, chemical synapses, local circuits, their plasticity, and the dynamics of large networks. They both remain to be connected to system-level neural correlates, such as temporal synchronization of transient neuronal assemblies. Network properties are, at least in principle, objectively measurable, yet they remain to be bridged to behavior or emotion, as subjectively acted or experienced.